Platform · Document intelligence
The documents that matter most are the badly behaved ones: 300-page filings, contracts with nested schedules, quality reports that are half spreadsheet and half scan. DataFact keeps their structure and remembers where every piece came from.
What gets stored
Reading a document is not the same as flattening it. DataFact records what each region on the page is, what it contains, and where it sits — so the memory keeps the shape a human would recognise.
Those regions are what an agent retrieves later, and what a reader is taken to when they want to check the source themselves.
| Tier | Q1 | Q2 |
| 4.2% | 4.8% |
Capabilities
Everything here exists because a real corpus broke a simpler approach.
Layout
Merged cells stay merged. Stacked headers keep their hierarchy. Multi-column pages read in the right order, and a scan sitting between two typeset pages is handled without dropping the rest of the document.
Chunks
Text, tables and figures are stored as distinct kinds of chunk. A table keeps its grid so it can be queried and rendered as a table; a figure keeps its image so it can come back inside an answer rather than as a filename.
Coordinates
Each chunk records the page and the region on that page it was taken from. This is what makes a citation actionable: the answer says page 12, and the viewer opens page 12 with the exact region outlined.
Review
Parsing is never perfect on the first pass over a difficult corpus. Chunks can be read, edited, split, merged, disabled or re-indexed in place, so one bad table does not mean re-ingesting a 400-page filing.
In an answer
When the useful part of a document is a diagram, a link to the file is not an answer. DataFact returns the figure itself, alongside the sentence that cites it, with the page it was cut from.
The paragraph or table cell the claim came from, quoted in full, with its page.
The rows themselves, rendered as a table with merged cells intact — not a screenshot.
The image, shown inline, with the region it occupies on the page marked in the original file.
Pick the document that has defeated other tools — the scanned appendix, the contract with nested schedules. We will parse it and show you what came back, chunk by chunk.
SaaS subscription or private deployment · No data migration either way